Abstract Background Computed Tomography Ventilation Imaging (CTVI) is an investigational technique that has its basis in functional lung avoidance radiotherapy. It offers a cost‐effective and accessible alternative to nuclear medicine imaging by generating lung ventilation maps from 4DCT or paired inhale/exhale breath‐hold CT (BHCT) scans. Despite over a decade of clinical validation, there is still no consensus on how algorithm parameters and patient‐specific factors influence CTVI accuracy. Further research is needed to understand CTVI's sensitivity to these variables and to standardize its implementation for clinical use. Purpose This study evaluates how key algorithm parameters and patient‐specific factors affect the accuracy of CTVI. Materials and methods CT ventilation images were generated from BHCT scans and compared to Galligas PET ventilation scans. The VESPIR toolkit was used to compute ventilation based on deformable image registration (DIR) evaluation of volume change (CTVI Jac ) or change in Hounsfield Unit (HU) value (CTVI HU ). CTVI accuracy was characterized as the voxel‐wise Spearman correlation (r S ) with Galligas PET. Algorithm parameters common to many CTVI implementations were investigated with a baseline determined from existing literature: lung segmentation threshold (−600 HU to −150 HU), DIR regularization parameter (λ = 0.05 to 100), and smoothing filter diameter (0 voxels to 9 voxels). Robust parameter ranges were defined as those yielding r S within 10% of the maximum cohort average observed through parameter variation, and no negative Jacobian values for the registration. Patient‐specific lung volume and density metrics were also analyzed to explain inter‐patient variability in CTVI accuracy. Results The correlation between CTVI and Galligas PET was demonstrated to be robust within identified parameter ranges: lung segmentation threshold −600 HU to −150 HU for CTVI Jac and CTVI HU , DIR regularization parameter (λ) 1.25 to 5 for CTVI Jac and CTVI HU , and smoothing filter diameter 0 to 9 voxels for CTVI Jac and 7 to 9 voxels for CTVI HU . No significant correlation was found between the accuracy of CTVI Jac and any patient‐specific lung volume or density parameters. Significant correlations were found between the accuracy of CTVI HU and the percentage change in lung volume during inspiration ( r = 0.72, p < 0.01) and the lung volume in the exhale phase ( r = −0.63, p < 0.01). The correlation between CTVI Jac and CTVI HU was found to be strongly correlated to CTVI accuracy. Conclusions CTVI accuracy was relatively stable across the range of parameter values tested with no strong indication of the need for patient‐specific parameter sets. Patient‐specific differences appear to be a driving factor for inter‐patient variability in CTVI accuracy as parameter selection alone was insufficient to explain the variability. The strong association of CTVI Jac and CTVI HU agreement and CTVI accuracy suggests that CTVI Jac and CTVI HU agreement is a useful predictor of CTVI accuracy and quality metric for parameter optimization.
Lim et al. (Fri,) studied this question.